To my best understanding this is basically doing PageRank but with votes taking the place of links—so a user’s outgoing trust is divided between other users in proportion to how much in total they’ve upvoted each one.
It seems to me this approach would be likely to strongly favor more prolific users, and I would guess that, even if outsiders didn’t agree with the core of prolific users, they would tend to see results weighted heavily towards those users and whoever those users most upvote.
I would much prefer an approach that compensated for this in some way.
It seems to me this approach would be likely to strongly favor more prolific users
That’s a very good point. I might upvote 20 out of 200 posts by a prolific user I don’t trust much, and 5 out of 5 posts by an unprolific user I highly trust. But this system would think I trust the former much more.
But then, just using averages or meduans won’t work, because if I upvoted 50 out of 50 posts from one user, and 5 out of 5 of another user, then I probably do trust the former more, even though they have the sma eaverage and median, 50 posts is a much better track record than 5 posts.
To my best understanding this is basically doing PageRank but with votes taking the place of links—so a user’s outgoing trust is divided between other users in proportion to how much in total they’ve upvoted each one.
I could well be wrong though, the documents include a low-level description of the algorithm if people want to check.
It seems to me this approach would be likely to strongly favor more prolific users, and I would guess that, even if outsiders didn’t agree with the core of prolific users, they would tend to see results weighted heavily towards those users and whoever those users most upvote.
I would much prefer an approach that compensated for this in some way.
That’s a very good point. I might upvote 20 out of 200 posts by a prolific user I don’t trust much, and 5 out of 5 posts by an unprolific user I highly trust. But this system would think I trust the former much more.
But then, just using averages or meduans won’t work, because if I upvoted 50 out of 50 posts from one user, and 5 out of 5 of another user, then I probably do trust the former more, even though they have the sma eaverage and median, 50 posts is a much better track record than 5 posts.